Steve Higgs
06/07/2012 9:40 AM
post93570
|
Using QNX4.25 and TenAsys eVM our application experiences many -nan float problems during math equations that are
correct and boundary protected and have correctly executed 100s of times before. In some cases, a -nan is created doing
the equation on the same numbers that just milliseconds before worked correctly.
Our gut is that these problems are related to the eVM environment (possibly some confict between the two systems).
However, we have not been able to prove this.
Has any one else seem similar problems or had similar experiences?
Also when the -nan is occurs, no signal (SIGFPE) is raised. We have used sprintf(some string,"%f", float value) and
string compare for the "-nan" to catch these. It appears that pending on where this trap code is placed, the problem
seems to move which leaves a gut feeling that it is timing related.
Any additional comments would be appreciated.
|
|
|